Google DeepMind has announced that AI agents found two flaws in legacy 1024-bit RSA encryption in a matter of weeks, showing how quickly artificial intelligence is now advancing the science of breaking old codes. The work, announced on Friday, November 6, 2026, was carried out by AI agents guided by human cryptographers. DeepMind stressed that modern 2048-bit and larger RSA keys are not affected, and that no live systems were touched.

Key Points

• Google DeepMind announced on Friday, November 6, 2026, that AI agents guided by human cryptographers found two weaknesses in legacy 1024-bit RSA encryption in weeks.
• The first is a signature-forgery method the team estimates is about 1,600 times faster than the previous best for short 32-byte digests.
• In a sandboxed test on an already-solved benchmark key, the agents' code reproduced an earlier forgery using 3.5 times less CPU time.
• The second is a batch-factoring algorithm called "Fan Factories" (working paper, November 5, 2026) that cuts the cost of factoring each additional 1024-bit key by about 30 times.
• Modern 2048-bit and larger RSA keys are not affected, and US federal guidance has disallowed 1024-bit RSA for new signatures since 2013 — though such keys persist in older hardware, long-lived certificates and embedded systems.

What Happened

The work was led by DeepMind researchers Dj Dvijotham and Abhradeep Thakurta, working with cryptographers Nadia Heninger of the University of California, San Diego, and Emmanuel Thomé of Inria, the French national research institute. DeepMind's vice president of science, Pushmeet Kohli, set out the findings on X, framing them as a demonstration of what AI agents can now do in cryptanalysis.

The first breakthrough concerns digital signatures, the mathematical seals that prove a message or a piece of software really comes from who it claims to come from. The team says its agents found a signature-forgery method about 1,600 times faster than the previous best technique for short 32-byte digests. In a sandbox test against a benchmark key that had already been solved, the agents' code reproduced the earlier forgery using 3.5 times less CPU time.

The second breakthrough concerns factoring, the hard mathematical problem at the heart of RSA security. RSA works because multiplying two large prime numbers is easy, while working backwards — factoring the product to recover the primes — is enormously difficult. The new "Fan Factories" algorithm cuts the cost of factoring each additional 1024-bit key by about 30 times, to roughly one fifty-eighth of a full factoring run once a shared precomputation has been done.

Background: Why 1024-Bit RSA Still Matters

RSA encryption, named after its inventors Rivest, Shamir and Adleman, has underpinned secure communication on the internet for nearly half a century, protecting web traffic, software updates and email by relying on the sheer difficulty of factoring very large numbers. The "1024-bit" figure is the size of the key: the larger the key, the harder it is to factor, and the safer the encryption. By modern standards, 1024-bit RSA is already considered weak. The US federal government has disallowed it for new digital signatures since 2013, and the industry standard for new systems is now 2048 bits or more.

Yet old keys do not simply disappear. They persist in older hardware, in long-lived certificates issued years ago and never replaced, and in embedded systems — industrial controllers, medical devices and network equipment that can stay in service for a decade or more. Replacing cryptography in such systems is slow and expensive, and sometimes practically impossible without replacing the hardware itself.

How the 1024-Bit RSA Encryption Flaws Work

The two advances attack RSA from different directions, and that combination is what makes the announcement notable. The signature-forgery method targets the process by which a digital signature is created and verified: a forgery lets an attacker create a signature that appears valid without holding the private key. For short 32-byte digests, the DeepMind team estimates its method runs about 1,600 times faster than the previous best approach, turning a slow, expensive attack into a far more practical one.

The "Fan Factories" algorithm takes a different route. Factoring a single 1024-bit key from scratch remains a heavy task, but the new method changes the economics when an attacker wants to break many keys: by doing a shared precomputation once, the cost of each additional key falls to roughly one fifty-eighth of a full run. Both results were produced by AI agents working with human cryptographers, not by AI alone. The agents explored ideas and tested approaches at a speed no human team could match, while the researchers set the direction and validated the mathematics.

Reactions

DeepMind was careful to put the results in context. The company stressed that modern 2048-bit and larger RSA keys are not affected, and that the tests ran only in a sandbox, on keys that had already been solved, with no live systems touched. The announcement is a research result, not a report of a breach, and no one has used these techniques against a real system. Nadia Heninger's involvement signals how seriously the work was taken.

Why It Matters

The announcement matters for three reasons. First, it compresses time: work that might once have taken a research team months or years was completed in weeks. Second, it changes the economics of attacking legacy systems, since "Fan Factories" makes bulk attacks cheaper — once the shared precomputation is done, each additional 1024-bit key costs a fraction of a full attack. Third, it is a reminder that the cryptography protecting the internet is not a fixed wall but a moving target.

What Happens Next

The "Fan Factories" working paper, dated November 5, 2026, is now available for other cryptographers to examine, and the research community will test and verify the claims in the weeks ahead. For organisations, the practical next step is an audit: any system still using 1024-bit RSA should be identified and scheduled for replacement. Longer term, the work feeds into the shift toward post-quantum cryptography, as tech investment keeps pouring into AI-era infrastructure.

Frequently Asked Questions

What did Google DeepMind announce?
On Friday, November 6, 2026, DeepMind said AI agents guided by human cryptographers found two weaknesses in legacy 1024-bit RSA encryption in weeks: a faster signature-forgery method and a cheaper batch-factoring algorithm called "Fan Factories".

Are my passwords and bank accounts at risk?
No. Modern systems use 2048-bit or larger RSA keys, and DeepMind says those are not affected. The flaws concern 1024-bit keys, disallowed for new US federal signatures since 2013.

How much faster is the new signature-forgery method?
About 1,600 times faster than the previous best method for short 32-byte digests; in a sandboxed test it reproduced an earlier forgery using 3.5 times less CPU time.

What is "Fan Factories"?
A batch-factoring algorithm described in a working paper dated November 5, 2026. After a shared precomputation, it cuts the cost of factoring each additional 1024-bit key by about 30 times.

Where are 1024-bit RSA keys still used?
In older hardware, long-lived certificates and embedded systems such as industrial controllers and network equipment, which can stay in service for many years.